IP Reachability Differences: Myths and Reali<es

Transcription

1 IP Reachability Differences: Myths and Reali<es Dr. Dan Massey Colorado State University He Yan, Ben Say, Dave Oko, Brendan Sheridan Colorado State University Dr. Christos Papadopoulos Dr. Dan Pei Colorado State University AT&T Research 1

2 Ideally, all ISPs provide you with reachability to same Internet In practice: Different ISPs can reach different portions of the Internet. In other words, there are some prefixes that can be reached via ISP A but cannot be reached by ISP B (a vice versa). The reasons why this occur are well known: Bogon filtering causes reachability differences for newly allocated prefixes. Bogon filtering matters, but the fact is that regional differences matter most and these differences may be political, economic, technical, or a combination of the above Bogon filtering and regional issues matter, but the real difference is due to various customer only services. Even if there are differences, there are very little (if any) production services on networks that are not globally reachable from most ISPs. 2

3 Our Objec<ves 1. Demonstrate reachability differences exist and quan<fy the extent of these differences. 2. Examine which conjectures (if any) provide a significant contribu<on to the differences. 3

4 Methodology Compare BGP Rou<ng Tables from 25 ISPs Data collected from Oregon RouteViews All peers send full tables of over 310K prefixes Studied from April 1 to April 20, 2010 Study Uses ISP0 As a Central Point of View ISP0 is a <er- 1 ISP Have addi<onal data on ISP0 opera<ons Compare each of the other 24 ISPs to ISP0 4

5 Types of Table Differences Exact Matches: ISP0 prefix appears ISPX table Ex: Both ISP0 and ISP2 have a route for /16 Covered Prefixes: ISP0 prefix does not appear in ISPX, but ISPX can reach all IP addresses Less Specific: /17 does not appear in ISPX, but /16 does appear in ISPX More specifics: /17 does not appear in ISPX, but /18 and /18 appear in ISP X Remainder represents poten<al reachability differences. 5

6 Exact Matches Number of Prefixes % of ISP0 prefixes also appear in ISPX ISP Index (not shown here, but 97.25% of ISPX prefixes also appear in ISP0) 04/15/2011 IP Reachability: Myths and Reality 6

7 4500 The BGP Table Differences Number of Prefixes On average, 3250 prefixes appeared in ISP0 tables, but did not appear in ISP22 tables ISP Index 04/15/2011 IP Reachability: Myths and Reality 7

8 Covered Prefixes 7,000 Number of Prefixes 6,000 5,000 4,000 3,000 2,000 Not Covered Covered Nearly all ISP0 prefixes that don t appear in ISPX are covered by other table entries in ISPX 1, ISP Index 04/15/2011 IP Reachability: Myths and Reality 8

9 Overall Reachability Differences 10,000 1,000 (Prefixes) Not Reachable Partially Reachable Number of Prefixes ISP Index 04/15/2011 IP Reachability: Myths and Reality 9

10 Overall Reachability Differences 1e+08 1e+07 (IP Addresses) Number of Destinations (IPs) 1e ,000 10,000 1, ISP Index 04/15/2011 IP Reachability: Myths and Reality 10

11 Our Objec<ves 1. Reachability differences do exist. Small percentage (less than 3%) of total routes. Can be up to 1000 prefixes and millions of IP addresses 2. Examine which conjectures (if any) provide a significant contribu<on to the differences. 11

12 Transient Changes Short Lived BGP Table Differences Fast learners: IPS0 learns the route first, eventually other ISPs catch up and also learn the route Slow to react: ISPX learned the prefix is unreachable, but it takes longer for ISP0 to withdraw the route. Long Term BGP Table Differences. For mul<ple days, ISP0 could reach the prefix but ISPX could not. 12

13 Transient Behaviors 2,500 2,000 1/3 of reachability differences persisted for entire study period. Number of Prefixes 1,500 1, Number of Days Appeared 1/3 of reachability differences are short lived 04/15/2011 IP Reachability: Myths and Reality 13

14 Bogon Routes ISPs Apply Policies to Filter Invalid Routes No universal agreement on what is invalid New Alloca<ons and Bogons Filter unallocated prefixes and update filter ager alloca<on Going back 7 months accounts for 43 prefixes. Going back years yields 295 prefixes. Management Strategies and RADb 92 prefixes with no entry in RADb 199 prefixes with no origin in RADb. Does produce reachability differences, but only explains a small percent of differences. 14

15 Regional Differences 2,500 2, prefixes from India, only unreachable from ISP 22. ISP 22 added these routes on day 13 of study. Number of Prefixes 1,500 1, /15/2011 Number of Days Appeared Bulgaria accounted for most prefixes (509) Great Britain was second highest (168) IP Reachability: Myths and Reality 15

16 Customer Services and Prefix Usage ISPs Provide Specialized Services to Customers Confirmed ISP0 has several such prefixes Most are covered prefixes, but did find some uncovered prefixes Accounts for 9 prefixes in our study. Are Services Running on the Impacted Prefixes? Scanned prefixes looking for common services Limited scan to /19 or more specifics Responding prefixes further checked for DNS entries Contacted Prefix Owners Responses include reports of reachability problems, blacklist issues, network tes<ng, and so forth Valid Global Services Running on Clear Majority of Prefixes 16

17 RouteViews Data in Real- Time 04/15/2011 IP Reachability: Myths and Reality 17

18 Conclusions Reachability Issues Do Exist Over 97% of prefixes appeared as exact matches But differences do occur and impact both research and opera<ons Common Conjectures: Transient issues, Bogons, Management, Regional Policies, Configura<on Errors, Special Services all proposed as explaina<ons All are valid, but none are dominant. 18

19 Ques<ons 19